* don't rearm after a fatal error before connection setting changed ( does not affect the normal route on non fatal error like newtork issue)
* serialize error correctly in database to ensure the "avoid a database write per auto-reconnect" corectly trigger
a lot of more cleanup to ensure that the connection is really dropped even if the host changed its id
* fix a mishnalding in the grace period before marking a pool disconnected . this will keep the ui in sync AND not redownload all the xapi object for a transient issue
* fix an error transmitted as raw, so the user was only seeing {}
e58587aa1d